The Flexible Specialization Path of the Internet Petros KAVASSALIS Ecole polytechnique (CRG), Paris & Institute for Computer Science (ICS), Crete Email: petros@rpcp.mit.edu William LEHR MIT & Columbia University Email: wlehr@rpcp.mit.edu Abstract ________________________________________________________________________ The Internet provides a new paradigm for the evolution of communication industries, one that facilitates the emergence of new types of flexible organization structures. This is due in part to the philosophy and technology behind the Internet Protocol (IP), and also to the current stage of industrial development and convergence of communications media. We argue that the forces that have heretofore compelled vertical integration of telephone and cable television apply with weaker force to the Internet. Moreover, the inherent design features of IP allow it to serve as a separating spanning layer that decouples the innovation and provisioning processes for applications and the underlying facilities infrastructure. This reduces the asset-specificity inherent in earlier infrastructure and service architectures and enables industry participants to more flexibly organize and re-organize firm-specific assets, resulting in a proliferation of viable business strategies. The fulfillment of this process is likely to be an industry of both integrated and nonintegrated firm types, coexisting and using new types of emergent wholesale markets (e.g., IP transport services) to share assets across infrastructure and service provider platforms. This paper makes the case for why the Internet paradigm is substantively different from earlier communications technologies by analyzing trends in the technology and industry structure, and the economic forces driving those trends. The research presented here reflects the further elaboration and synthesis of work presented in Lehr (1998), Kavassalis, Bailey, and Lee (1998), and Kavassalis and Lehr (1998). Introduction1 ________________________________________________________________________ The Internet offers the promise of a new industrial paradigm for our information infrastructure. This may be contrasted with the organizational principles of traditional telecommunication networks in which applications are closely coupled with the supporting infrastructure and in which the service providers control network intelligence through hierarchical switching and complex signaling protocols. As the Internet matures, it is undergoing rapid change as a consequence of: __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 1 of 23 i) Infrastructure liberalization and deregulation; ii) Growth in the demand for and supply of IP-based services and products; and, iii) The need and desire to provide integrated voice, video, and data services on a single network platform. These changes have profound implications for the future governance structure of the Internet industry. Building on earlier work,2 we examine the forces for vertical integration of network facilities ownership and retail services that are characteristic of network industries and consider these in light of the new opportunities for more disintegrated industry structures enabled by the Internet and IP-technologies. We explain how IP-technology may facilitate the disintegration of traditional network ownership structures, giving rise to new business models. However, while we believe that the Internet will eventually emerge as a significant force for a more open, less integrated industry structure; in its current stage of development, the forces compelling increased integration remain strong. The Internet is a network built largely on top of the existing telecommunications infrastructure. Internet Access Providers (IAPs) such as, for example, America Online in the United States or TOnLine in Germany, and private corporate and academic networks offer local access to the Internet to individual users and computers. Internet Service Providers (ISPs) such as UUNET, MCI, and Sprint provide connectivity between the edge users (IAPs and private networks) and the Internet cloud which is constituted from many interconnected ISP networks.3 In most cases, the local transport facilities connecting individual users to the IAPs and the backbone wide area ISP networks rely on traditional telecommunications facilities and services (e.g., leased Lines, Frame Relay, SMDS, ATM, or SONET) to interconnect their Points of Presence (POPs) at Network Access Points (NAPs). The current allocation of ownership and control responsibilities over these co-specialized assets (POPs, NAPs, IAP and ISP networks, and the underlying physical infrastructure) is not always clear. In some cases, ISPs lease essential facilities from infrastructure providers (mostly, from the telecommunications companies); in other cases, the ISPs are vertically integrated. End-to-end vertical integration wherein a single firm retains ownership and managerial control over the entire value chain from basic infrastructure through to delivering retail services to the final customer is common among telecommunications carriers. Some analysts (Gong and Srinagesh, 1997) see an increased trend towards vertical integration occurring in the Internet services industry. The large market shares of such integrated carriers as MCI, Sprint, AT&T, and the European Telecom Operators, as well as such proposed mega-mergers as WorldCom and MCI, seem to reinforce this impression. The key point of this paper, however, is that there is something fundamentally different about IP and the Internet. The Internet Protocol weakens traditional forces for vertical integration, thus allowing for a less integrated industry structure to emerge. This argument is developed in subsequent sections as follows: (1) Something is different about IP: The Internet Protocol provides a spanning layer in the form of a bearer service that facilitates the evolution of an open network architecture that is both technology independent and application blind. __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 2 of 23 (2) Therefore, IP supports vertical disintegration: The Internet Protocol supports less integrated organizational forms that permit the separation of the provision of end-user services and the underlying infrastructure. (3) However, strong forces for vertical integration still apply. The same forces that have helped drive integration in other network industries such as telecommunications and electric power transmission apply to the Internet, albeit in a weaker (and weakening) form. These include: • Scale and scope economies (network, retail, and common costs) • Coordination costs (metering costs and Quality of Service needs) • Product differentiation and price discrimination (recovery of capacity costs, avoidance of destructive price competition) • Innovation and strategic positioning • Market power (4) The current trend towards consolidation and vertical integration may be explained as the result of adaptive learning and “coordinated expectations”. Traditional forces for vertical integration and mutual observation drive incumbent firms to pursue vertical integration strategies. However, the emergence of new types of “specialized” Internet providers such as Qwest, Savvis, InterNAP, and Level3 demonstrate that alternative models of organization structure are feasible. In the new world enabled by the Internet, vertically integrated and non-integrated, specialized carriers will compete. 1. IP is different ________________________________________________________________________ The Internet is a network of architecturally different networks exchanging IP traffic. The Internet protocol (IP) is the lower layer of the open Internet protocol suite that serves as a common denominator or interface among heterogeneous infrastructure technologies (Ethernet, Frame Relay, ATM, ISDN etc.). Also, IP enables heterogeneous applications to request network service in a fashion that is independent from the underlying physical network technologies. In that way, the current version of IP serves as an elementary bearer service for many interoperable applications (email, ftp, web, etc.) that can be supported by multiple network technologies (Ethernet, frame relay, ATM, ISDN, etc.). This facilitates the separation of infrastructure provisioning from service offering. With the addition of several extensions, IP may be evolving into a more complex interface capable of providing Quality of Service (QoS) guarantees which will allow the Internet to reliably support such real time services as telephony and video conferencing. This extension of IP capabilities helps drive the process of horizontal integration (or convergence) of all media (voice, audio, video, animation, and data) into a common but heterogeneous network and terminal infrastructure – fueling the process of convergence. In addition, as we will demonstrate later, the diffusion of IP as a common interface, or spanning layer, has important implications for the evolution of markets and industry organization. __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 3 of 23 1.1 IP is an efficient spanning layer As David Clark (1994) has pointed out, the IP protocol serves as a spanning layer that allows translation to occur between a range of infrastructure technologies. This facilitates the harmonic coexistence of different network technologies and the easy integration of new ones. As a result, above the spanning layer, interoperation between applications is easier to achieve because it does not require an understanding of how the underlying infrastructure either works or evolves. In general, a spanning layer is helpful because it provides generic (default) rules that abstract from the functional details of particular situations, which permits easy translation and/or interoperation between a range of context-specific rules that may be applied either above or below the spanning layer.4 The ability of IP to fulfill the role of a spanning layer follows from its design. Digital networks are organized into a hierarchy of layers wherein each layer is responsible for a specific set of functions. IBM's Systems Network Architecture (SNA), the 7-layered Open System Interconnection Model (OSI), and the collection of Internet protocols are some of the better known examples of network protocol suites.5 Each layer in these architectures is responsible for a specific set of functions. For example, in the OSI model, the physical layer at the bottom is responsible for the transmission of raw bits. Protocols at this level are tied to the specific media over which the transmission occurs (e.g., copper wire, fiber optic cable, etc.). The second or data link layer is responsible for organizing these bits into structured messages. The third or network layer is responsible for routing messages between the source and destination. The fourth or transport layer provides end-to-end transport services; and so on, up to the seventh application layer which is the one end-users interact with. Each lower layer provides service to the next higher layer by exporting a clean interface (i.e., a protocol) which defines the exchange rules between layers. The two best known Internet protocols are the Transmission Control Protocol (TCP), which provides transport layer services, and the Internet Protocol (IP), which provides network layer services. The IP protocol resides above the physical layer, and hence, by design, is essentially independent of the wires. Thus, IP can support communications over many types of physical media. The TCP protocol was designed to support the transmission of delay tolerant applications such as those that comprise most of the traffic on the Internet today (i.e., email, telnet, ftp, and web browsing). Additional protocols and enhancements are required to allow the Internet to reliably support such delay-sensitive, real-time applications as telephony and video conferencing. But again, the evolution of the design at the upper network layers relies on IP, not directly on the underlying infrastructure technologies. 1.2 Separating services and infrastructure facilities Drawing on Clark’s analysis, Messerschmitt (1996) discusses the importance of spanning layers in effectively isolating the design of the horizontal layers above and below. Similarly, in a system composed of multiple components, a spanning layer supports exchanges between components at different layers, which may evolve independently. The Microsoft Windows Operating System provides a clear example of an effective spanning layer. The operating system/application program interface (even though a private standard) exports a clean interface to computer applications, which allows them to share the functionality provided by the underlying hardware layers while remaining ignorant of the hardware’s functional details. This permits independent design and widespread __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 4 of 23 interoperability of PC hardware and software applications. The transmission standards NTSC, PAL and SECAM (these are collective standards) that guaranteed interoperability between appliances (television sets) and applications (television broadcasts) offer another good example of a spanning layer. These standards facilitated the rapid expansion of the industry (although they may have slowed the pace of technical innovation). In communication networks, it is possible to regard the infrastructure facilities and the services that use those facilities as distinct, in a fashion analogous to the distinction between hardware and software applications in computers. Historically, however, communication networks have evolved as single application networks (telephony or cable television) managed by a monopoly provider. Thus, this distinction has not been as apparent and the potential role of a spanning layer separating facilities and services has been largely unrecognized and unexplored. In the digital era and in the face of successive waves of deregulation, the potential for such separation has become more apparent. The proliferation of services (e.g., fax, data services, etc.) and the emergence of wholesale markets for bulk switched and dedicated transport (e.g., leased lines, virtual private networks, etc.) further highlight the potential for separating control of the innovation and provisioning processes for services and network facilities. Although perhaps not readily apparent, a spanning layer has always existed in communication networks but closely linked to particular infrastructure technologies. In telephony, it has evolved from the initial 4 KHz analog voice channel to the 64 Kbps bit pipe (reinforced progressively by additional switching and control functions, and with the diffusion of X.25 and ISDN architectures). The most recent incarnation of this spanning layer is offered by ATM, which accomplishes two main functions (Kavassalis, Lee and Bailey, 1998): i) ATM defines a basic set of services supported by the infrastructure or substrate technologies (or bitways); and, ii) ATM translates application requests for network services to the substrate technologies. While the ATM interface offers a mechanism for supporting a diverse array of applications, it is still firmly linked to the underlying facilities infrastructure. Changes in the underlying infrastructure affect the spanning layer. In contrast, the Internet spanning layer offered by IP is superior because it allows the spanning layer to move "out" of the bitways or wires, thus becoming independent of the substrate technologies. This allows a more complete decoupling of infrastructure facilities and service innovation and provisioning. This helps sustain and promote the emergence of new organizational forms for service providers that stand in contrast to the traditional model of vertical coupling of infrastructures provision, applications, and services within the same firm that has heretofore characterized communication networks.6 (Figure 1 provides a picture of this evolutionary process.) __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 5 of 23 Figure 1: IP as a spanning layer 2. IP supports vertical disintegration ________________________________________________________________________ This separation of services and infrastructure technologies implies a radically new approach to designing networks, which also impacts the evolution of industry structure. In Sustaining a Vertically Disintegrated Network through a Bearer Service Market, Kavassalis, Lee and Bailey (1998) develop this argument by defining two different technological trajectories (in the sense of Dosi, 1982) or approaches to technological development. They used metaphors from the transportation industry to explain the difference between the two approaches as they apply to the evolution of the Internet.7 These two approaches, referred to as the Telecom and the Internet trajectories, represent competing paths for industry evolution. Each implies a fundamentally different cognitive view of what the network should do. Each is discussed separately in the following two sub-sections. 2.1 The Telecom trajectory The Telecom trajectory is the path followed by the traditional telecommunications operators that provide the facilities that support the Public Switched Telecommunication Network (PSTN). Traditional telecommunication networks bundle together bitways and applications in such a way that it is cognitively difficult to distinguish between the conception of an application and its delivery. Applications are strongly tied to particular substrate technologies. For example, in the traditional PSTN, data applications were provided on X.25 circuit switched networks, while “integrated” voice and data applications were to be provided via ISDN bitways. When the applications and the network are closely coupled in this way, network innovations require changes both in the core of the network and at the end nodes. Moreover, innovation at the applications level __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 6 of 23 is limited to what has been conceived and designed ex ante by the engineers of the infrastructurefacilities owners. ATM technology represents the ultimate stage for this cognitive model for network design. Telecom Operators8 have proposed ATM as an integrating platform which may accommodate a diverse mix of traffic types (voice, data, video) within the same single high-speed substrate (Iwata et al., 1995). This offers one strategy for putting an end to the period of single-application, coupled infrastructures (i.e., separate telephone, cable television, and data networks). However, in exactly the same way as with older telecom technologies, the ATM architecture locates the spanning layer within the bitways (i.e., the constitution and transport of ATM cells is decided at the lower, infrastructure, layers of the digital hierarchy). While ATM may be capable of supporting a far broader range of applications than the traditional PSTN (including connectionless data) and ATM supports a higher degree of modularity for network functions, it still limits the range and flexibility of application innovations.9,10 This occurs because these are limited to those that are compatible with the set of capabilities embedded in the substrate ATM bearer technology. 2.2 The Internet trajectory In contrast, the Internet approach is very different. IP is located “in the middle” of the digital hierarchy, on top of the bitways.11 IP acts as a separator, rather than an integrator, offering a spanning layer capable of isolating the design of layers below and above it. Consequently, IP supports a clean separation between infrastructure facilities (below) from service offerings (above).12 By defining a technology-independent interface to the infrastructure facilities (i.e. a variable size datagram) and then exporting blind functionality for applications development, the Internet Protocol is capable of supporting multiple applications (e.g. e-mail, login, ftp, web and now, telephony, multicast etc.). Furthermore, with the addition of QoS attributes, the suite of Internet protocols may evolve into a bearer service for an even larger family of applications, including real-time ones (i.e., delay intolerant services such as telephony).13 The major difference between the Telecom and Internet approaches is the flexibility inherent in the latter with respect to the use of network resources. In the Internet, applications may be supported over multiple infrastructure substrates (including ATM, frame-relay, Ethernet, etc.). Similarly, infrastructure technologies do not pre-specify the development of new applications. As a result, variety in the design and innovation of applications is reinforced, making the product line openended. Thus, IP drives the communication network industry to flexible specialization development paths (Piore, 1989, 1995). In contrast, the telecom approach follows the more typical pattern of modern industrial systems. In short, the Internet is to the PSTN what containerization was to the rail roads. Containerization revolutionized intermodal transport by providing a universal interface over different physical media and vehicles (rail-railroad cars, highways-trucks, water-vessels, airairplanes and helicopters, etc.). The Internet can serve a similar function for communication networks. __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 7 of 23 2.3 Trajectories and economic structures Each of the above approaches has different implications for the economics of optimal firm structure.14 The telecom approach is associated with the mass production of a standard service (i.e. telephony) that is supplemented with incremental and progressive variation through the addition of data and integrated services (but within a space with boundaries predetermined by the design of the infrastructure technologies themselves). As we explain further in Section 3 below, this is associated with the model of a large integrated corporation. The traditional telecommunications company, which is fully integrated from infrastructure to final services, provides an example of such a firm. It is internally organized in specialized divisions supporting autonomous product lines and managed by a very hierarchical structure. Management strives to ensure coordination over the complex technical system in order to transform high fixed costs into low unit costs.15 In this model, there were three forces encouraging vertical integration: i) Technological economies from internal coordination of local access infrastructures with central switches and long distance lines; ii) Transaction economies arising from asset-specificity related to sunk investments in infrastructure facilities; and, iii) Succesful innovation in a multi-component system. Progressively, as the telecomunications industry has become more competitive, new suppliers have appeared at the margin in the form of (i.) facilities-based competitors, (ii) downstream providers of a final service (for example, value added services, provided over wholesale facilities that are leased from the vertically integrated suppliers), and (iii) simple resellers (and discounters) of long distance telephony. However, the industry structure has not changed significantly with respect to control over the network value chain. Incumbent firms, holding residual rights of control (Grossman and Hart, 1986) over the relationship-specific assets (i.e. transport links, access switches and signaling network), retained their ability to apply vertical restraints to independent providers.16 With the emergence of the Internet, this began to change. We associate the Internet with the flexible specialization model which faciltates the emergence of a new institutional structure. This model supports successive independent markets, wherein providers of essential upstream assets produce intermediary inputs for downstream firms. We may think of the Internet as a two-stage production system: upstream infrastructure facilities providers offer network capacity to Internet Service Providers (ISPs), which in turn, may sell IP-packet transport either directly to the customers (firms or individuals) or to other ISPs, IAPs or service integrators. This new industry structure is illustrated in Figure 2 in contrast to the PSTN model that preceded it. __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 8 of 23 Figure 2: Independent markets for bandwidth and IP connectivity The emergence of a new intermediary stage offering IP connectivity while using input provided by various physical facilities (leased lines or “cloud’ technologies such as Frame Relay, SONET, or ATM) supports the emergence of less aggregated organizational forms. This has had profound implications for industry structure. First, it has given rise to new types of communication firms: Internet Service Providers. The traditional carriers failed to recognize the importance of the Internet and the implications of TCP/IP for network design and services. This occurred, in part, because the Internet represented a radically different paradigm for how networks should be designed and because it implied a very different path for industry evolution. This new paradigm did not coincide with the traditional carriers’ prior learning or experience.17 This created a window of opportunity for a whole new class of firm types such as the Internet Service Providers (ISPs) and Internet Acces Providers (IAPs) that emerged to sell IP connectivity in wholesale and retail markets, respectively. Second, the Internet’s packet-switched architecture facilitates interconnection and interoperability across heterogeneous networks (Clark, 1988). This capability affects investment incentives and ownership structures since it makes physical network assets less co-specialized (i.e., with the addition of an IP spanning layer that is “outside of the wires”, network facilities that were insalled to support one type of application, can be used to support multiple applications). By taking advantage of this flexibility, the emerging ISPs and other service integrators using IP technologies have been able to compete successfully with the historical infrastructure facilities owners (i.e. telecommunications companies). Third, the emergence of additional wholesale markets for intermediary inputs has made it feasible for less-integrated types of firms to emerge that specialize in only a subset of the industry value __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 9 of 23 chain. A similar process occurred in the computer industry with the emergence of specialized component manufacturers, value-added resellers, and other types of firms.18 The growing number of ISPs and flexible service integrators illustrate this phenomenon, in spite of predictions that the Internet would be dominated by a few large vertically integrated firms that would be able to foreclose smaller, less-integrated participants from the market. Furthermore, new governance forms allow for “selective intervention” (i.e. specialization at one stage with selective vertical expansion at the neighboring stages).19 Evidence for such business models is accumulating from companies such as Savvis and Qwest.20 Finally, the Internet trajectory has important implications for the future growth of the industry. Following Piore’s (1990, 1992) description of mechanisms for growth in mass production and flexible specialization models, it is reasonable to suppose that a multi-stage production system with markets for intermediary products (i.e. bandwidth, IP connectivity, bearer services, etc.) and firms with various degree of vertical integration, is likely to be better suited to support applications which are: i) Dynamic and changing in response to users requirements, rather than changing because of the requirements or limitations of the underlying infrastructure. The Internet implements applications in the end-nodes which support easier evolution and change; ii) Semantically independent from network technologies, i.e. provided across heterogeneous infrastructure networks. This allows uncertainty with respect to application needs or demand to be partially decoupled from uncertainty about the capabilities or supply of different types of infrastucture (cable TV, copper wire, wireless). The Internet’s fast diffusion is fueled by the simultaneous diffusion of a variety of applications that are also subject to rapid diffusion and change (Shuster, 1998). In this environment, communications services (i.e. the bundling of applications with networks in a customer-package) can be readily customized, and hence, specialized to address the needs of specific customers. Service providers need to be able to reconfigure the functional elements of their offerings to adapt to the rapidly changing output requirements and emergence of new markets. For some functional elements (for example, middleware services or portal access for electronic commerce offerings), network structures (Piore and Sable, 1984; Powel, 1991) may be the appropriate governance structure. For others, such as bandwidth, IP connectivity, or a bearer service (bundled or not with middleware services), the market appears to guarantee the required organizational flexibility (the pure market or the infusion of market governance into hierarchies, a point to which we will come back later). 3. The recent revival of the “hierarchies” ________________________________________________________________________ Given the arguments in the preceding sections, how are we to understand the current trend towards vertical integration via mergers (GTE-BBN, WorldCom-UUNET) or internal expansion (MCI, Sprint, European Telecoms)?21 According to the influential trade journal, Communications Week International, firms that want to succeed in this sector need to be big because in order to win “you are going to need the capital and network infrastructue” of a worldwide company.22 Therefore, the only strategic dilemma apparently facing would-be Internet firms is whether to buy or build. __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 10 of 23 Evidence from the industry press23 suggests that for both, Telecommunications Operators and Internet Providers, incentives for vertical integration were initially driven by the need to assure markets for outputs or adequate supplies of essential inputs. Traditional telecommunications facilities owners perceived the Internet as a new market for data services and, albeit after initial delay, have entered the ISP and IAP markets successfully (i.e., forward integration). On the other side, the peculiar US Internet interconnection model (peering), has encouraged ISPs to build their own national infrastructure (i.e., backward integration) in order to meet the “peering requirements” (i.e. 45 Mbits links and access to three NAPs) for interconnection with the five biggest US backbones. But this is only a conjectural interpretation. Further, one may ask whether other more fundamental trends apply. The increasing number of mergers and other vertical movements suggests that some of the functions performed within the emerging complex institutional structure may appear as a return back to the tightly integrated, hierarchical, corporate organization. Is this a permanent shift towards a stable oligopoly or something else? In order to address this question, we need first to understand the determinants of the recent vertical integration movements and second, in the next section, to examine the relationship of this recent revival of vertical integration and the flexible specialization model. As discussed by Lehr in Understanding Vertical Integration in the Internet, there are strong incentives to vertically (and horizontally) integrate in communication network industries, as is evident from an examination of the older telephone and cable television networks. In the following sub-sections we review briefly two sets of factors that encourage vertical integration: scale and scope economies and organizational economies, and explain why these forces are weaker in the Internet than in traditional communication networks. 3.1 Scale and scope economies Scale and scope economies exist whenever there are large fixed or shared costs. The cost structure of communication networks includes many fixed or shared costs at each level within the value chain. These exist with respect to network investment and operating costs, with respect to retaillevel costs, and with respect to overhead, or common costs. Network facility investments provide the first source of scale and scope economies. Networks are sized to handle the expected peak capacity and so most of the costs of investment and network operations do not vary with the level of network usage. The same capacity can be used to support multiple services and so the capacity costs are shared. This creates opportunities for increasing returns to scale. For example, the network operations expenses (planning, maintenance, and management)24 do not increase proportionally with network size. There are also significant increasing returns associated with installing transport facilities. Because it is only slightly more expensive to install ten fibers than it is to install one fiber once you have prepared the conduit, carriers typically install excess capacity to meet future demand.25 In the Internet world, however, these network increasing returns are reduced. First, Internet computer technologies are more modular and scalable. This means that capacity can be added in smaller increments. Second, __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 11 of 23 Internet traffic is more heterogeneous which increases opportunities for statistical multiplexing. And, third, IP packet routing increases opportunities to share capacity. Scale and scope economies are also available at the retail-level. These costs also include significant shared and fixed costs. For example, the costs of establishing a brand image via national advertising do not vary with the amount of traffic handled. Product innovation and marketing management costs and the costs of setting up a billing system are also mostly fixed. Moreover, there are scope economies associated with marketing multiple services to each customer (e.g., one stop shopping). The effect of the Internet on retail-level scale and scope economies is ambiguous. On one hand, customer uncertainty with an emerging technology may increase the importance of brand-related reputation costs; while on the other hand, the trends toward modular and open systems and the growth of outsourcing alternatives (e.g., third party billing) makes these costs more variable. A final source of cost-based scale and scope economies is associated with corporate overhead. This includes a portion of the costs of corporate management and for such headquarters activities as payroll and accounting and regulatory affairs. The last can be quite expensive, serving as an entry barrier for small firms. If the Internet leads to reduced regulatory compliance costs, it will reduce the scale and scope economies associated with common costs. Scale and scope economies also arise on the demand-side due to network externalities (i.e., it is more valuable to be connected to a larger network). Vertical and horizontal integration permit firms to realize the benefits of these externalities, albeit to a lesser extent in an Internet world. However, because the Internet promotes ubiquitous interconnection, smaller networks can share the benefits of universal termination supported by the larger networks.26 3.2 Organizational economies In addition to the scale and scope economies discussed above, there are a number of other economic reasons for why vertical integration is generally more efficient for communication networks. These include the following: 3.2.1 Metering costs and Quality of Service One of the justifications for not employing usage-sensitive pricing in the Internet today is due to the costs of metering traffic. This includes both creating the appropriate infrastructure (i.e., modifying existing routers and servers) and the administrative overhead to meter traffic. By giving rise to homogeneous networks, managed by a single operator, vertical integration eliminates the need to meter traffic within the core of the network, thereby reducing transaction costs. Also, related to metering, an important driver for end-to-end integration may be the need to provide integrated, Quality of Service differentiated services. This may become a powerful incentive for integration in a homogeneous entity in order to assure service reliability and efficient allocation of resources in a network with multiple classes of service. It is not possible to support such differentiated end-to-end services over the public Internet today, but extensions to the suite of Internet protocols are under development that are expected to remedy this situation.27 __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 12 of 23 3.2.2 Product differentiation or price discrimination The need to recover the fixed and partially sunk costs of constructing network facilities provides a powerful inducement for facilities-based providers to integrate forward. This allows them to differentiate their products and offers better account control to permit price discrimination via value-added services. Because there is a substantial risk that available industry network capacity will exceed demand and because short-run incremental costs are significantly less than long-run average costs (which include fixed and shared costs), it may be difficult for facilities-based providers to avoid aggressive "Bertrand-like" price competition (Gong and Srinagesh, 1996, 1997). In order to price-discriminate, facilities-based providers will have a strong incentive to offer bundled services (i.e., one stop shopping and services that bundle transport with value-added features such as enhanced billing, new features, etc.). Creative bundling will facilitate a wider range of discount programs that can be used to more narrowly target customer groups. Moreover, onestop shopping bundles offer opportunities to offer forward-discounts (i.e., rebates to customers who stay with you, or volume discounts over multiple services) that reduce customers' incentives to switch to a competing carrier. The impact of these forces is to provide an incentive for facilities-based providers to integrate downstream into additional retail-level markets such as Internet access or hosting services. However, this does not preclude these facilities providers from participating also in wholesale markets which will facilitate competition from non-integrated retail-level service providers. 3.2.3 Innovation and strategic positioning In technologically advanced markets, the need to vertically integrate in order to fill a new opportunity (e.g., no current supplier upstream, no distribution channel downstream) is often an important motivation for vertical integration. Also, vertical integration may be needed to develop complementary skills.28 This does not seem to provide a valid rationale for integration of local and wide area facilities providers, because these services already exist and the skills associated with each are not substantially different. However, this does provide an incentive for facilities providers to integrate into IAP or ISP services. Integration in this case may provide new learning opportunities and thus, result in an expansion of firm-specific competencies. 3.2.4 Market power Vertical integration is often pursued or considered as a strategy to protect, exploit, or extend market power. Because this is harmful to competition, antitrust policy scrutinizes the effects of vertical mergers for their effect on the competition. To the extent that current efforts to promote competition in all telecommunications service markets, especially local telephone services, are successful, the market power of established incumbents will be reduced. This will reduce their ability and incentive to use vertical integration to pursue market power goals. The Internet is playing an important role in how these markets __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 13 of 23 develop (e.g., competition from Internet Telephony service providers for fax and traditional voice telephony customers), however, the elimination of traditional industry bottlenecks (e.g., local access facilities) and the avoidance of new potential bottlenecks (e.g., perhaps associated with domain naming services) cannot be assured. This means that vertical integration remains for anticompetitive purposes remains an important potential motivation. Unfortunately, a more complete discussion of this issue is beyond the scope of this work. 3.3 Interpretations The preceding discussion helps explain why we are observing a wave of vertical (and horizontal) integration associated principally with the facilities-based telecommunications providers that have traditionally dominated communication services (e.g., Bell Atlantic and GTE, SBC and Ameritech, AT&T and British Telecom, AT&T and TCI, etc.). The strategy of “getting bigger to survive” is driven by two goals: i) Facilities-providers seeking the optimal scale or size to fully exploit scale and scope economies to achieve a high-level of capacity utilization and low unit costs. ii) The need to create a proprietary and homogeneous space within the big Internet wherein the firm can efficiently allocate resources (and price accordingly) and better appropriate any value from innovation. This trend, while currently quite strong, is very much in the tradition of industry evolution in the past. This is just the model of a Chandler-Williamson infrastructure corporation (Williamson, 1985) producing everything internally and more efficiently than markets: as the industry matures, large companies integrate into successive stages of the value chain to capture ever larger scale and scope economies, to better coordinate adjacent production stages, to better implement/disseminate bestindustry practices, and to respond to competitive threats. In the Internet, however, as we noted above, there are a number of forces that make it viable for a wide array of alternative, less-integrated strategies to succeed. In an Internet world, the large integrated communications company will confront difficult challenges: i) As communications applications become independent from the underlying technologies, they will become even more dynamic, even relative to the increased dynamism fueled by media convergence. Product design processes will change and market boundaries will continue to be agitated by innovation. Rather than transitioning to a new stable equilibrium, the Internet will help make rapid change a permanent market condition. Therefore, firms that pursue a low cost strategy by offering a static “same-old” internal product line will suffer and new organizational learning is required.29 (Tushman, 198X). ii) Integration between Internet and Telecom “core competencies” has proven to be more difficult in practice than it might first appear.30 These difficulties have been observed also in other sectors of the industry (e.g., when mergers between hardware and software firms or computer and communications firms founder). Core competencies are not easily enhanced through synergies when capacity provisioning and retail-services are combined. __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 14 of 23 iii) Vertically integrated companies will face sharp competition not only from other members of a closed set of large firms but also from new entrants, flexible service aggregators, or specialized providers that focus on particular stages of the network services value-chain. 4. External economies of scale and metastability ________________________________________________________________________ The key proposition of this paper is that the Internet facilitates more flexible specialized strategies. This is because the Internet increases the importance of market coordination relative to organizational coordination in the value-creation process for the communications industries. Vertical integration to form big corporations, driven by the desire to exploit internal scale economies remains a strong trend; however, we are also now seeing the emergence of new types of companies that specialize in providing services specialized to only a subset of levels within the value chain (e.g., bandwidth markets, IP services, generic or specialized communications services). Both forms of organization -- integrated firms (selectively or entirely) and firms more devoted to flexible specialization (relying on markets to source inputs) may prove successful strategies. The coexistence of these strategies may give rise to multiple industry equilibria, distinguished by varying degrees of market share between integrated and non-integrated firms. Multiple equilibria are often observed in industries that are growing as rapidly as the Internet, both with respect to demand and the range and size of supply.31 In the case of the Internet, however, there are two important reasons why this multiplicity is likely to remain a permanent feature. That is, we are unlikely to see the industry settle into a structure consisting either of vertically integrated firms (perhaps the traditional incumbents repositioned or perhaps replaced by new entrants that have replaced the incumbents but have adopted a similar integrated structure) or non-integrated firms. Rather, we will see both strategies co-existing. First, the growing importance of external economies of scale. These arise when suppliers spread the use of specialized machinery across multiple customers to facilitate high utilization.32 These external economies help explain Qwest’s business model of developing extensive infrastructure to provide just-in-time services to other facilities-based and non-facilities based service providers. The emergence of these Carrier’s Carriers will encourage the participation of integrated carriers in wholesale bandwidth markets, further reinforcing these markets. These, in turn, will create opportunities for the emergence of additional types of firms such as Saavis and InterNAP. These latter firms are offering private super-NAP services by managing multiple connections to parallel backbones. Thus, the existence of the external infrastructure economies creates new business opportunities.33 As these markets mature, the cost to both integrated and non-integrated carriers of provisioning bandwidth for services should narrow. Consequently, the modern communications firm’s efficiency may derive more from the flexible utilization of externally provided network resources than from the internal development of large infrastructure plants. Even within vertically integrated firms, market mechanisms may improve the efficiency of coordination across business units (e.g., infrastructure facilities and customer services departments).34 Again, it is the organizational separation of service offerings from infrastructure facilities (through the IP bearer service) which allows external economies obtained at one level in the value chain35 to be transferred within the __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 15 of 23 whole system. Sturgeon’s (1997) research on the US automobile, computer, and other industries characterizes the emergence of turnkey production networks as a new model of American industrial organization. The term describes a system of “specialized and highly capable merchant suppliers that provide the industry (i.e. the brand name firms) with a functionally coherent set of co-modified production services (essentially manufacturing capacity).” The second force making it more likely that the multiplicity of equilibria will remain a permanent feature is the rapid pace of innovation fueled by the Internet. This supports fundamental cognitive diversity wherein industry participants develop different perceptions of “what the problem is.”36 One’s worldview depends on one’s perspective. If the world is changing rapidly (because of the Internet) and this pace of change is likely to continue (again, in part, because of the Internet), then it becomes less likely that a common worldview will emerge. It’s the old problem of the blind men trying to comprehend an elephant by touching only a small part, made more difficult because the elephant is itself continuously changing. For traditional carriers, familiar with the imperatives of exploiting scale and scope economies to lower costs and self-provisioning end-to-end services to protect specialized assets and assure coordination, vertical integration is the preferred solution. Other ISPs may focus on other cost or demand related problems, reflecting a different cognitive model for how communications networks ought to evolve. The increased potential for cognitive diversity may induce inconsistency regarding the interpretative devices (Arthur et al, 1997) by which agents anticipate the future. This may induce an interesting pattern of evolution at the macro-level (Darley and Kauffman, 1997): when the system approaches a stable state (for example, the vertically integrated firm equilibrium), it tends to break to a disordered stage from which other business models may emerge (metastability). If this is the case, the Internet may be following organizational patterns similar to the “wheel of retailing”(Brown, 1988). As the retailer matures and grows, it becomes less efficient and its costs/prices rise, creating a niche for new discount retailers. __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 16 of 23 Bibliography ALLEN D. (1997), ARTHUR B., DURLAUF S. & LANE D. (1997), The Economy as an Evolving Complex System II, Addison-Wesley, Reading (Mass.). BOHLIN E. & LEVIN S. (1998), Telecommunications Transformation, Technology, Strategy and Policy, IOS Press, Amsterdam. BRADASH J. & ECCLES R. (1989), Price, Authority and Trust: From Ideal Types to Plural forms", Annual Review of Sociology, no. 15, pp. 97-118. BROWN S. (1988), The Wheel of the Wheel of Retailing, Intl. J. Retail, no 1, pp. 16-37. CHANDLER A. (1977), The Visible Hand: The Managerial Revolution in American Business, Harvard University Press, Cambridge (Mass.). CLARK D. (1988), The Design Philosophy of the DARPA Internet Protocols, Computer Communication Review, no 4, pp. 106-114. CLARK D. (1994 & 1995), Interoperation, Open Interfaces and Protocol Architecture, MIT Lab. of Comp. Science (LCS), mimeo. COHEN X. & LEVINTHAN X. (1990), CSTB (1994), Computer Science and Telecommunications Board, National Research Council, Realizing the Information Future: The Internet and Beyond, National Academy Press, Washington DC CSTB (1996), Computer Science and Telecommunications Board, National Research Council, The Unpredictable Certainty: Information Infrastructure Through 2000, National Academy Press, Washington DC. DARLEY V. & KAUFFMAN S. (1997), Natural rationality, in Arthur, Durlauf & Lane (eds). DOSI G. (1982), Technological paradigms and technological trajectories, Research Policy, no, pp. DOSI G. (1988), DOSI G. & EGIDI M. (1991), "Substantive and procedural rationality", Journal of Evolutionary Economics, no.1, pp. 145-168. DOSI G. & MARENGO X. (1994), FURGESON C. & MORRIS C. (1993), Computer Wars, Times Book, Random House, New York. GONG J. & SRINAGESH P. (1996), "Network Competition and Industry Structure", Industrial and Corporate Change, no. 5, pp. 1231-1241. GONG J. & SRINAGESH P. (1997), The Economics of Layered Networks, in McKnight and Bailey (eds). GROSSMAN S. & HART O. (1986), The Costs and Benefits of Ownership : A Theory of Vertical and Lateral Integration, Joournal of Political Economy, no 94, pp. 691-719. HOLLAND J. (1995), Hidden Order, Addison-Wesley Pub., Reading (Mass.). __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 17 of 23 IWATA A., MORI N., IKEDA C. & SUZUKI H., ATM connection and traffic management schemes for multimedia networking, Communications of the ACM, 38, pp. 72-89. KAVASSALIS P. & SOLOMON R. (1997), "Mr Schumpeter on the Telephone: Patterns of Technical Change in the Telecommunications Industry before and after the Internet", Communications & Strategies, IDATE, no. 26, pp. 371-408. KAVASSALIS P., LEE T. & BAILEY J. (1998), Sustaining a Vertically Disintegrated Network Through a Bearer Service Market, in Bohlin & Levin (eds). KAVASSALIS P. & LEHR W. (1998), Forces for Integration and Disintegration in the Internet, Communications & Strategies, IDATE, no 30, pp. KATZ M. (1989), Vertical Contractual Relations, in Schmalensee & Willig (eds). LEHR W. (1998), Vertical Integration and Internet Industry Structure: An Application of the Pricing Taxonomy, MIT Internet Telephony Consortium Working Paper. LEIDA B. (1998), A Cost Model of Internet Service providers: Implications for Internet Telephony and Yield Management, MS Thesis, MIT. LESOURNE J. (1991), Economie de l’ordre et du desordre, Economica, Paris. MCKNIGHT L. & LEIDA B., Internet Telephony: Costs, Pricing and Policy, Chapter submitted to the volume of selected papers from the 1997 Telecommunications Policy Research Conference, D. Waterman and J. Mac-Kie Mason (eds).. McKNIGHT L. & BAILEY J. (1997), Internet Economics, MIT Press, Cambridge (Mass.). BAILEY J., McKNIGHT L & SHARIFI H. (1998), Critical Business Decisions for Internet Services, Paper presented at the Resources Management Association International Conference, May 17-20, Boston. MESSERSCHMITT D. (1996), "The Convergence of telecommunications and computing: What are the implications today?", Proceedings of the IEEE, no. 8, pp. 1167-1185. NELSON R. (1992), "The Role of Firns in technical Advance: a Perspective from Evolutionary Theory", in Dosi et al., Technology and Entreprise in a Historical Perspective, Clarendon Press, Oxford. NOHRIA N. & ECCLES R. (eds) (1992), Networks and Organizations: Structure, Form and Action, Harvard Business School Press, Boston. PERRY M. (1989), Vertical Integration: Determinants and Effects, in Schmalensee & Willig (eds). PERRY M. & GROFF R. (1982), Vertical Integration and Growth: An examination of the Stigler story, Bell Labs Economic Discussion, paper no. 257. PIORE M. (1989), Corporate Reform in American Manufacturing and the Challenge to Economic Theory, MIT Sloan School of Management Working Paper. PIORE M. (1990), The Economic Revival of the 1990s: Technology, Organisational Structure and Competitive Strategy, OECD Working Paper. PIORE M. (1992), Fragments of a Cognitive Theory of Technological Change and Organizational Structure, in Nohria & Eccles (eds). PIORE M. (1995), Beyond Individualism, Harvard University Press, Cambridge (Mass.). __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 18 of 23 PIORE M. & SABEL C. (1984), The Second Industrial Divide, New York: Modern Library. POWEL W. (1990), Neither Market nor Hierarchy: Network Forms of Organization, Research in Organizational behavior, no 12, pp. 295-336. SCHUSTER M. (1998), Diffusion of Network Innovation: Implications for Adoption f Internet Services, MS Thesis, MIT. SCHMALENSEE R. & WILLIG R. (1989), Handbook of Industrial Organization, Elsevier, North Holland. STIGLER G. (1951), "The division of labor is limited by the extend of the market", Journal of Political Economy, no. 59, pp. 185-193. STURGEON T. (1997), Turnkey Production Networks: A New American Model of Industrial Organization, MIT Center for Technology, Policy and Industrial Development Working Paper. TEECE D. (1986), Profiting from Technological Innovation, Research Policy, no 6, pp. 285-305. TEECE D. (1988), TEECE D., WINTER S. & DOSI G. (1994), TENNENHOUSE D., LAMPSON B., GILLETT S. & KLEIN J. (1995), Virtual Infrastructure: Putting Information Infrastructure on the Technology Curve, MIT Lab. of Computer Science, TM-253. TUSHMAN m. (198X), WILLIAMSON O. (1991), Comparative Economic Organization: The Analysis of Discrete Structural Alternatives, Administrative Science Quarterly, no ?, pp. 269-296. ZENGER T. & HESTERLY W. (1997), The Disaggregation of Corporations: Selective Intervention, Highpowered incentives, and Molecular Units, Organizational Science, no 3, pp. 209-222. __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 19 of 23 __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 20 of 23 Notes 1 The authors would like to acknowledge the financial support of the Massachusetts Institute of Technology's Internet Telephony Consortium (MIT ITC) and many helpful comments and suggestions on earlier drafts from David Clark, Lee McKnight, Michael Piore and the other participants in the MIT ITC; and from participants in the EuroCPR98 conference organized by ENCIP, including Alain De Fontenay (de Fontenay, Savin & Kiss). Remaining errors are due to the authors. 2 This paper builds on work that was prepared on behalf of the MIT ITC. It synthesizes and extends two earlier papers, one by William Lehr entitled Vertical Integration and Internet Industry Structure: An Application of the Pricing Taxonomy and a paper by Petros Kavassalis, Thomas Lee, and Joseph Bailey entitled Sustaining a Vertically Disintegrated Network Through a Bearer Service Market. An earlier, very preliminary version of this paper, has been published in Communications & Strategies as “Forces for Integration and Disintegration in the Internet” (issue 30, 1998, 2nd Quarter 1998, IDATE, Montpellier). 3 A note on terminology: We introduce the designation of IAPs and ISPs to distinguish between the activity of providing end-user access to the Internet (IAP) and the activity of proving connectivity and backbone transport services in and across the Internet (ISP). Both are commonly referred to as ISPs in the trade press. 4 The discussion here draws upon Holland’s (1995) “internal models” of adaptation for complex systems. 5 For a discussion of the design of these and other network protocols see Andrew Tannenbaum, Computer Networks, Second Edition, Prentice Hall, Englewood Cliffs, NJ, 1988. 6 As Tennenhouse at al. (1995) point out, “this coupling is a consequence of a historic model in which suppliers bind each type of information to a physical or electromagnetic medium to which it is specifically suited and for which a distribution channel can be tailored. Newspapers articles are bound to newsprint for distribution by paper boys and telephone calls are bound to 4 kHz signals...” In the history also, this coupling went together with the emergence of the Chandlerian merger-generated firms, in both the railroad and telecommunication sectors. 7 They trace the evolution of rail roads to the emergence of containerization and its impact on intermodal transport services. This provides a useful case study from which to draw analogies for the coevolution of the Internet and telecommunication networks. 8 See Kavassalis and Solomon (1997) for a discussion of how ATM applies in an Integrated Broadband Network (IBN) approach. 9 When ATM bitways support directly applications, knowledge of service semantics resides primarily in the switches, “commanded” by a centrally controlled SS7 signaling network, so variety at the upper layers is still conditioned by technical change at the infrastructure layers. 10 Using Piore’s (1989, 1995) description of the on-going transformation of an industrial production system, Kavassalis et al (1998) discuss the evolution of the PSTN from an initial mass-production model, custom-tailored to support a single service (i.e., telephony), towards flexible mass production (or closed flexible specialization) practices. 11 At the narrow point of the so-called hourglass picture (CSTB, 1994; Clark, 1995). 12 See also CSTB (1996). 13 While it is possible to support Internet Telephony today over a managed Intranet, the quality of service when offered over the public Internet is still not deemed to be adequate. More advanced services that are delay-intolerant will require extensions to the Internet protocol suite. 14 As Nelson (1992) points out, co-evolution of technologies and organizations is a general pattern in the transformation of the industrial system. Piore considers organizations as a derivative concept from technologies, the later being defined, in reference to cognitive theory, as “the embodiment of certain concepts or conceptual frameworks in terms of which we think about transforming resources” (1992:440). According to Piore, different technological trajectories generate particular institutional structures. 15 To use Chandler’s (1977) The Visible Hand terminology __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 21 of 23 16 For an overview on the possibilities for vertical contractual restraints, see Katz (1989). 17 New technological trajectories bring in new learning models for extracting value from technology (Dosi, 1988; Teece, 1988), requiring the development of new core competencies (Teece et al., 1994; Lesourne, 1991). Incumbent firms do not always respond with success to the challenges posed by an emerging trajectory. As Cohen and Levinthal (1990) point out, the capacity of firms to evaluate and utilize outside knowledge is a path-dependent process, and in this sense it is a function of firm’s level of prior related knowledge. It is the prior related knowledge which confers them an ability to recognize the value of new information, assimilate it, and apply it to commercial ends”, an ability that constitutes a firm’s absorptive capacity. From this point of view, a firm’s “absorptive capacity” bounds the problemsolving space -- in that it relates to both “being good at certain things” and “being good at learning certain things” (Dosi and Marengo, 1994) and not others. 18 See also Allen (1997) 19 Williamson (1991) uses the term “selective intervention” to define combinations between markets and hierarchies (arguing that even if possible such combinations should be inefficient). In this paper, we reserve the term for mixed-strategies: specialization at an intermediary production stage and limited expansion to targeted downstream markets. In our view, this is the case of Qwest, which is exploring its strong position in the bandwidth market to expand towards end-user markets such as Internet telephony, where knowledge on the efficient allocation of bandwidth may offer a competitive advantage. This definition of “selective intervention” is close to what neoclassical economists call “vertical expansion”. According to Perry (1990), this term describes vertical integration “which occurs as a result of internal growth of the firm, creating its own new subsidiaries in the neighboring stages”. 20 Indeed, one can see the confirmation of these trends in the emergence of companies as Qwest and Level3 (bandwidth market) or Savvis and InterNAP (ISP market). Qwest is the best example of a specialized provider (with a very large and fast fiber network across the US, using both ATM and IP technologies), offering abundant bandwidth in the wholesale (carriers and ISPs) and retail markets (i.e. directly to business customers or individual consumers -- IP telephony). Savvis and InterNAP are the more prominent examples of a new kind of ISP, local aggregators. They aggregate traffic from other ISPs or web-based providers (as Point Cast or CDNow) and then, forward it quickly to the big national backbones (MCI, Sprint, UUNET, ANS etc.) through private NAPs (a sort of “aggregate hubs”, concept similar to airlines hubs). Private NAPs-based companies obtain much better performance and reliability because, i) they avoid highly congested public NAPs and MAEs (where ISPs interconnect each other) and, ii) do not rely on one only backbone. 21 The mega-merger between Worldcom and MCI and the recent AT&T-British Telecom partnership to develop a global international IP-based network, providing integrated end-to-end services, are two recent examples of the consolidation trend. 22 Communications Week International, June 30, 1997. 23 See, Telecommunications and the Cook Report on the Internet. 24 It is likely that the returns to scale for network management first decrease significantly (i.e., very small networks are easy to manage, but quickly becomes more difficult as they grow larger (extreme example, no problem with single computer becomes big problem with network of two computers), then increase over some range, and then eventually decrease again (i.e., it is possible for a network to be too large). 25 These increasing returns appear less important today because (1) network costs represent a smaller share of total costs; and, (2) the advent of such new technologies as Wave Division Multiplexing which make it relatively easy to significantly expand the capacity of fiber facilities at a low incremental cost later. 26 Currently, large backbone carriers exchange traffic using “bill and keep” arrangements. These presume that the costs of termination are minimal or that traffic is balanced. This raises the interesting question of who should be allowed to peer with who, which is ultimately a question of what smaller networks should pay for universal termination services. Currently, a number of large backbone carriers refuse to peer with smaller networks and charge those networks capacity and usage sensitive interconnection fees for transport services. The implications of these new arrangements for interconnection are difficult to evaluate (Bailey et al., 1998). But if these agreements are not used to impose proprietary interconnection constraints over the Internet, they may induce incentives for more __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 22 of 23 efficient allocations of resources through the pricing system. 27 The IntServ and DiffServ working groups in the Internet Engineering Task Force (IETF) are working on developing technologies to support quality-of-service differentiation within the Internet. 28 Teece (1986) has argued that innovators may have strong incentives to integrate into related stages (backward, forward, lateral) since such integration helps to better capture value in “weak appropriability regimes”. To give an example, an Internet telephony software company may have to integrate backwards to produce board-level product if no supplier can be found to provide. Or, more probable, upstream facilities providers or ISPs may need to integrate forward to enter the Internet Telephony market (Qwest and Deutsche Telecom follow currently this strategy for developing Internet Telephony services). 29 Recent re-engineering plans undertaken by the major European Telecommunications Operators are very demonstrative in this regard. 30 The press frequently reports on serious misunderstandings after the merger between traditional telecommunications operators and new Internet companies (for the difficulties of the merger between GTE and BBN, see Financial Times of 10.xx.1997). 31 See Stigler (1951); Perry and Groff (1982). 32 We thank T. Sturgeon for interesting discussions on this subject. In addition, see Bradash and Eccles (1989). 33 The network unbundling aspect of the Internet with respect to infrastructure facilities is perhaps the most important and easiest to comprehend impact of the Internet, however, it affects other business functions as well (e.g., customer service, billing, marketing, Web hosting, Web portals, voice mail, etcetera). Markets beget markets that support the emergence of the virtual corporation in competition with integrated corporations. A similar process was fueled in the personal computer industry by the existence of the S100 bus architecture that helped promote the emergence of markets for an ever finer class of sub-components for personal computers, so IP helps fuel unbundling networks at ever finer levels of aggregation. This fuels the 34 This is a more general trend, described by Zenger and Hesterly (1997) in terms of a parallel movement towards the infusion of some hierarchy into market exchange and some market governance into hierarchy. 35 The infrastructure level and the cloud, create external economies which may be explored by IP bearer service providers (or Service Integrators) and edge ISPs, respectively. 36 These lines draw upon Dosi and Egidi (1991) and Arthur, Durlauf and Lane (1997). __________ Kavassalis & Lehr Flexible Specialization Path of the Internet Page 23 of 23